Bug 171390 – B3::FoldPathConstants does not consider the fall through case for Switch

• Status: RESOLVED
• Resolution: FIXED
• Component: Java

[Saam Barati]

Say you have a B3 switch that's the moral equivalent of this C this:

```
switch (v) {
1: doSomething; break;

2: doSomethingElse;  break;

3:
default: BB that uses v
}
```

we'll rewrite the BB to replace "v" with the number 3. This is obviously wrong.

Note that the phase already handles IR for:
```
switch (v) {
1:
2: BB;  break;

default: break;
}
```

We are smart enough to not replace "v" inside BB with either "1" or "2".

[Filip Pizlo]

Wow, that's glorious.

[Saam Barati]

Created attachment 308444 [details]
patch

[Saam Barati]

Comment on attachment 308444 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=308444&action=review

> Source/JavaScriptCore/b3/B3FoldPathConstants.cpp:109
> +                if (switchValue->hasFallThrough())
> +                    targetUses.add(switchValue->fallThrough(block), 0).iterator->value++;

If we wanted to, we could be a bit niftier here:
We could check if any target is zero, and if so, we could check if targetUses(fall through) === 0, and if both are true, we could say that in the default block, the switch value is nonzero (obviously only if default block has a single predecessor). I'm not sure if it's worth bothering though.

[Build Bot]

Attachment 308444 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/b3/testb3.cpp:10784:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10785:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10786:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10786:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10787:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10787:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10788:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10788:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10789:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10789:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
Total errors found: 10 in 5 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Filip Pizlo]

Comment on attachment 308444 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=308444&action=review

R=me without that if.

>> Source/JavaScriptCore/b3/B3FoldPathConstants.cpp:109
>> +                    targetUses.add(switchValue->fallThrough(block), 0).iterator->value++;
> 
> If we wanted to, we could be a bit niftier here:
> We could check if any target is zero, and if so, we could check if targetUses(fall through) === 0, and if both are true, we could say that in the default block, the switch value is nonzero (obviously only if default block has a single predecessor). I'm not sure if it's worth bothering though.

VALIDATE(value->as<SwitchValue>()->hasFallThrough(valueOwner.get(value)), ("At ", *value));

Therefore, you don't need to check if hasFallThrough().  It must hasFallThrough().

[Saam Barati]

Created attachment 308448 [details]
patch for landing

[Build Bot]

Attachment 308448 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/b3/testb3.cpp:10784:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10785:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10786:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10786:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10787:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10787:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10788:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10788:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10789:  Consider using CHECK_EQ instead of CHECK(a == b)  [readability/check] [2]
ERROR: Source/JavaScriptCore/b3/testb3.cpp:10789:  Tests for true/false, null/non-null, and zero/non-zero should all be done without equality comparisons.  [readability/comparison_to_zero] [5]
Total errors found: 10 in 5 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[WebKit Commit Bot]

Comment on attachment 308448 [details]
patch for landing

Clearing flags on attachment: 308448

Committed r215908: <http://trac.webkit.org/changeset/215908>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.